PowerOn/service-llm-private
3
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

neues text model

Browse source
This commit is contained in:
ValueOn AG 2026-02-06 13:35:00 +01:00
parent a96da92704
commit d322281cda
2 changed files with 10 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
app.py
View file

@ -187,11 +187,11 @@ rateLimiter = RateLimiter(
# Model mapping: external name -> internal Ollama model name
# Production models (optimized for 32GB RAM server):
# - deepseek-ocr: 3.34B params, 8K context, ~6.7GB RAM
# - qwen2.5vl:7b: 8.29B params, 125K context, ~6GB RAM
# - granite3.2-vision: 2B params, 16K context, ~2.4GB RAM
# - qwen2.5:7b: 7.6B params, 128K context, ~4.7GB RAM (Text)
# - qwen2.5vl:7b: 8.29B params, 125K context, ~6GB RAM (Vision)
# - granite3.2-vision: 2B params, 16K context, ~2.4GB RAM (Vision)
MODEL_MAPPING = {
"poweron-ocr-general": "deepseek-ocr",
"poweron-text-general": "qwen2.5:7b",
"poweron-vision-general": "qwen2.5vl:7b",
"poweron-vision-deep": "granite3.2-vision",
}
@ -536,7 +536,7 @@ async def _analyzeDocument(
# Model-specific context lengths (actual model limits)
modelContextLengths = {
"deepseek-ocr": 8192, # 8K context
"qwen2.5:7b": 32768, # Use 32K (model supports 128K but RAM limited)
"qwen2.5vl:7b": 32768, # Use 32K (model supports 125K but RAM limited)
"granite3.2-vision": 16000, # 16K context
}

10
templates/index.html
View file

@ -824,11 +824,11 @@ Falls ein Feld nicht erkennbar ist, setze den Wert auf null.</textarea>
ollamaModel: 'granite3.2-vision'
},
{
name: 'poweron-ocr-general',
displayName: 'PowerOn OCR General',
description: 'Text-Extraktion / OCR (deepseek-ocr)',
name: 'poweron-text-general',
displayName: 'PowerOn Text General',
description: 'Text-Verarbeitung / Analyse (qwen2.5:7b)',
isVision: false, // Text model - no image required
ollamaModel: 'deepseek-ocr'
ollamaModel: 'qwen2.5:7b'
}
];
@ -1174,7 +1174,7 @@ Falls ein Feld nicht erkennbar ist, setze den Wert auf null.</textarea>
// Model-specific context lengths
const modelContextLengths = {
'deepseek-ocr': 8192,
'qwen2.5:7b': 32768,
'qwen2.5vl:7b': 32768,
'granite3.2-vision': 16000
};